denser v0.2.0-alpha.1 — Codex context-savings audit

This alpha turns denser into a behavior-fidelity gate for LLM context changes and publishes its first result that clears the project's end-to-end savings bar.

Recorded Codex result

Under Codex CLI 0.147.0, gpt-5.6-sol, medium reasoning, and three randomized trials per case:

• Release-operation decisions: 27/27 in each profile; 20,294.11 -> 18,154.00 input tokens/call (-10.55%)
• Automation permission routing: 15/15 in each profile; 20,619.00 -> 18,434.00 input tokens/call (-10.60%)
• 84/84 calls passed; 0 operational errors; 0 transport fallbacks

The first strict run exposed one missing output-contract behavior. text-only/v1 fixed that boundary, then the complete 84-call audit was rerun.

Included

• denser audit: paired behavior replay with a required known-bad control
• Codex CLI text-only/v1: removes unused capabilities for pre-bundled text decisions
• reproducible benchmark, complete per-call evidence, source hashes, and case study
• Python package version 0.2.0a1

Scope

text-only is not a coding mode. Use it only when the complete input is already in the request. Tasks needing files, shell, network, plugins, apps, skills, or memory must keep standard.

Verification

GitHub CI passed on Python 3.10, 3.11, 3.12, and 3.13, plus the package smoke test.

• denser-0.2.0a1-py3-none-any.whl SHA256: 6aa83f6b65ee119e6276103b608eb1986b2f5a35db9320b41b6449ce5833144e
• denser-0.2.0a1.tar.gz SHA256: 2e4e820e5c68bc716d9ef4f2adf7a926f283393fdde40501a89a326895d81ff5

See the Codex text-only case study.